AI Coal Maceral Analysis With Noise Image Removal

The patent replaces the manual mechanical inspection process with an automated image processing system using deep learning algorithms. The system captures images of coal macerals and uses neural networks to automatically classify them, eliminating the need for manual visual inspection by operators while maintaining high classification accuracy and significantly reducing analysis time.

The patent creates digital copies (images) of the maceral samples and uses these copies for automated analysis through deep learning models. Instead of requiring operators to directly examine physical samples, the system processes high-resolution images, allowing multiple analyses simultaneously and eliminating operator-dependent variability.

The patent replaces the human expert's visual inspection capability with an automated deep learning system. The neural network is trained on labeled maceral images to learn classification patterns, thereby encoding expert knowledge into the algorithm. This allows non-expert operators to achieve expert-level classification accuracy without requiring extensive training or experience.

The system performs self-learning through the deep learning training process, where the model automatically improves its classification capability by learning from labeled training data. The algorithm independently identifies patterns and features without requiring continuous human guidance or adjustment, making the system self-sufficient and easy to operate.

The patent transitions from one-dimensional point-counting analysis to two-dimensional image-based analysis. Instead of examining individual points sequentially, the system captures and analyzes the entire maceral structure in two dimensions using image processing, allowing simultaneous evaluation of multiple features and significantly increasing analysis throughput while maintaining consistency.

The system performs preliminary image capture and preprocessing before classification, creating a complete digital record of the sample structure. This preliminary action allows the classification algorithm to simultaneously evaluate all relevant features without sequential point-counting, thereby maintaining standardized classification while dramatically improving productivity.